Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Sat, 24 Mar 2012 05:45:50 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: -DDEVICE_IS_CPU for OpenCL builds

On Fri, Mar 23, 2012 at 09:41:22AM +0100, magnum wrote:
> On 03/23/2012 09:29 AM, Samuele Giovanni Tonon wrote:
> > Btw i was thinking that maybe it could be best to move out rawsha1
> > opencl as possible format: CPU will always be faster so why keep a dupe ?
> > Are there other opencl formats that could be moved out (in the unused
> > dir) so we can focus on less but good formats?
> 
> I'm not so sure about that. How about users that have OpenCL but not
> SSE2?

Yes, and also how about running both the CPU and the GPU formats at once
(in different sessions), to make use of both devices?  Even if the GPU
code is badly inefficient, it may provide some minor speedup compared to
using the CPU only.

> Besides we might find ways to change the john core to support fast
> OpenCL formats better. By the way the latter could be a GSoC task if it
> isn't already.

There is a "JtR: GPU for fast hashes" task, but I doubt that any student
who is not already involved with our project will propose any specific
enhancements to JtR now.  I have my own thoughts on such enhancements,
and I'm afraid that I'm the one to find time and implement them - not
only for the GPU stuff, but also for better OpenMP scaling with fast
hashes (that's what I'd test the formats interface changes on).

Alexander

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.